It seems that, generally, “fakeraid” cards are not well supported, and Linux’s software raid is just as good.
Bonnie benchmarks on a Areca 11XX SATA RAID card (128MB cache) comparing a RAID10 config (4xMaxtor 300GB 7200rpm 16MB) in hardware mode versus Linux software RAID. Bandwidth tests match up with a bit more CPU usage (not surprising). Slightly larger gaps in file creation/search/deletion as those tests hit 99%+ CPU usage in every case which decrease available CPU for handling RAID requests.
Hardware Software
---------- ----------
Seq Out Char 45M/s 91% 43M/s 91%
Seq Out Block 124M/s 41% 124M/s 49%
Seq Out Rewrite 47M/s 9% 47M/s 12%
Seq In Char 50M/s 87% 51M/s 95%
Seq In Block 118M/s 10% 121M/s 12%
Rnd Seek .60T/s 0% .64T/s 0%
Seq Create 2.3T/s 99% 1.7T/s 99%
Seq Delete ----- -----
Rnd Create 2.5T/s 99% 1.6T/s 99%
Rnd Delete 9.6T/s 99% 8.8T/s 99%
My experience with hardware LSI MegaRAID SCSI controllers show even better numbers using Linux software RAID. Unfortunately, all my systems using LSI controllers are in production at this time so I can’t break apart the drives to run benchmarks.
Comments
Add Comment